Hardware Accelerated Hybrid Classifier for MCI Detection

Mild Cognitive Impairment (MCI) is progressive in the world population of 55 million, and 60–70% above the age of 60 are reported to have Alzheimer’s Disease(AD). AD is the seventh causing the highest number of deaths. People find it difficult in the process of remembering without depending on others. Remembering the bills paid, eating, and recognizing their family members will be difficult. But in mild AD, people can be diagnosed. In moderate stage of AD, the person will require much care as the symptoms worsen over time. The person with moderate Alzheimer’s might feel angry, irritated because of damage caused to the nerve cells. There might be a lot of mood swings and mentally challenging situations. There might be trouble in controlling the bowels and bladder. Their sleep patterns might be disturbed which thereby increases the brain damage. In the late stage of AD, the symptoms are severe. A person with AD lose the ability to talk and control movement. They will require full-day assistance from others. There will be changes in the physical activities like walking, swallowing, and difficulty in communicating with others. There are various tools available for diagnostic purpose. This work consists of both software and hardware Field Programmable Gate Array (FPGA) implementation using RNN, CNN, and hybrid algorithms. The collected MRI data is trained, tested, and fed into Xilinx Vivado toolbox. The result of the Xilinx is then loaded to spartan-6 FPGA board. The results of both hardware and software are compared for accuracy. Recurrent Neural Network with Random Forest algorithm has the highest accuracy of 88.9%. The accuracy is more efficient in hybrid algorithms when compared to standalone algorithms.